The invention relates to the field of robots and associated command systems. More precisely, the subject of the invention is a method for detecting the collisions of a robot with its environment.
The invention applies advantageously to any type of manipulator robot, in particular serial manipulator robots but also manipulator robots with tree-like structure or with closed kinematic structure.
In recent years, numerous architectures of lightweight robots have been proposed. The avowed aim of this new generation of robot is to interact with human operators while carrying out a task. The robot and the operator therefore share the same work space, and this implies new problematic issues pertaining to the safety of the operator. Indeed, movements of the robot with large dynamic range can notably limit the access zone near the robot.
A general problematic issue relating to manipulator robots pertains to the detection of collisions between the robot and its environment. Indeed, with the aim of improving the operating security of the robot, it is important to be able to rapidly detect collisions between the robot and its environment so as to minimize the possible damage by applying suitable post-impact strategies.
Known collision detection algorithms are usually based on the comparison of measurements with a model, making it possible to create a signal called a “residual” and which constitutes an image of the collision. The mathematical modeling of the system never being perfectly representative of the real behavior of the robot, the residual is marred by errors, and the detection strategies must impose the use of safety margins (manifested in practice by thresholds) in order to be robust in relation to these errors. This avoids the occurrence of false alerts. But on account of these conservative margins, the robot loses its sensitivity to collisions.
The present invention is thus aimed at improving the collision detection schemes based on the generation and the evaluation of a residual while removing or limiting the impact of system modeling errors.
As will be developed subsequently, the schemes for detecting collision between a robot and its environment are generally designed according to a two-step approach: a first step of generating the residual, a term which contains the item of information associated with the collision phenomenon, and a second step of evaluating this residual which consists notably in taking a decision regarding the occurrence of a collision as a function of the value of the residual.
Concerning the first step of generating the residual, several approaches are inspired by the scientific field of automation. It is notably possible to cite document [1] which utilizes a Kalman estimating filter, whose estimated quantity is used as residual. Document [2] describes a more specific approach which arises from fault detection. Article [3] describes a diagnosis observation method used on a robot arm in order to generate the residual. In the field of robotics, the most widely used residual generating strategy consists in comparing a measurement of the articular torques τ with their estimation {circumflex over (τ)} deduced on the basis of the inverse dynamic model of the robot. This technique has been improved by using the equation for the generalized momentum of a robot in articles [4] and [5], thereby avoiding the on-line calculation of the acceleration and reducing the influence of the measurement noise. Another strategy described in [6], based on filtering the torque, makes it possible to avoid the calculation of the acceleration. All these schemes are based on a model of the robot, and this therefore implies sensitivity to modeling errors.
Concerning the second step of collision detection, the evaluation of the residual makes it possible to take into account the uncertainties of the model. It consists in verifying that the residual is below a certain threshold. If the residual exceeds this threshold, the algorithm detects collision. This threshold makes it possible to establish a margin so as to render the scheme insensitive to modeling errors and to avoid false alerts. The simplest technique, described for example in document [5], consists in comparing the residual with a constant or static threshold, which therefore represents the maximum modeling error. This scheme considerably reduces the sensitivity of the robot to a collision since a static threshold does not make it possible to discriminate the modeling error related variations of the residual.
More advanced strategies use dynamic thresholds, whose level is adapted on-line. It is possible to cite article [3] in which an adaptive threshold is generated with the aid of techniques based on fuzzy logic. This solution is fairly complex to put in place, since it requires data gathering in order to develop the fuzzy logic laws. In articles [6] and [7], several dynamic thresholds are proposed, taking into account the parametric uncertainties. These solutions are based solely on the inverse dynamic model of the robot which, even if the parameters of the model are estimated perfectly, does not constitute sufficient modeling. Indeed, it does not take into account certain physical phenomena specific to manipulator robots, in particular lightweight serial robots, such as the phenomena of flexibilities.
The invention proposes a method for detecting collisions which takes into account the errors in modeling the residual and makes it possible to solve the limitations of the known solutions set forth hereinabove. The modeling errors, which comprise at one and the same time uncertainties related to the parameters of the model of the robot and also the non-parametric uncertainties related to factors outside the chosen model, are identified and filtered in order to generate a filtered residual. A dynamic threshold is determined on-line so as to adapt the detection criterion to the uncertainties related to modeling errors.
The application of the method according to the invention makes it possible to preserve good sensitivity to collisions while decreasing the false alarm phenomena related to the modeling errors of the inverse dynamic model used to generate the residual.
The performance of the method according to the invention is notably improved for robots subject to large dynamic ranges in terms of velocity or acceleration.
The subject of the invention is a method of detecting collision between a robot composed of a plurality of bodies linked together by at least one articulation and its environment, said method comprising the following steps:
According to a particular aspect of the invention, said adaptive threshold T is composed of the sum of several dynamic terms TΔ1, TΔ2, TΔ3 each equal to an item of information regarding the parametric uncertainties between said model and the real behavior of the robot, said parametric uncertainties being related to a different variable, characteristic of the state of said robot from among the following variables: the position, the velocity or the acceleration of a fixed point of the robot, or a linear or nonlinear function of one of these variables or of a combination of these variables.
According to another particular aspect of the invention, said adaptive threshold T furthermore comprises a static term Tstatic configured so as to be greater than a measurement noise level.
According to another particular aspect of the invention, the filtering step is carried out by means of a recursive least squares algorithm.
This filtering step can comprise the following sub-steps:
According to another particular aspect of the invention, an additional step of temporal filtering is applied to the filtered residual for each of its components. The additional step of temporal filtering can be a step of root mean square calculation.
According to another particular aspect of the invention, the step of determining at least one first dynamic term TΔ1, TΔ2, TΔ3 of the adaptive threshold T is carried out by means of a recursive least squares algorithm.
The step of determining at least one first dynamic term TΔ1, TΔ2, TΔ3 of the adaptive threshold T can comprise the following sub-steps:
According to another particular aspect of the invention, an additional step of temporal filtering is applied to said dynamic term TΔ1, TΔ2, TΔ3 of the adaptive threshold T for each of its components.
The additional step of temporal filtering can be a step of root mean square calculation.
According to another particular aspect of the invention, the step of comparing the filtered residual with the adaptive threshold T so as to deduce therefrom the existence or otherwise of a collision comprises the following sub-steps:
According to another particular aspect of the invention, the step of generating the residual consists of the following sub-steps:
According to another particular aspect of the invention, the parametric uncertainties related to the articular acceleration of the robot are uncertainties regarding the inertia matrix of the robot.
According to another particular aspect of the invention, the parametric uncertainties related to the articular velocity of the robot are uncertainties regarding the matrix of the centrifugal and Coriolis terms of the robot and/or regarding the viscous frictions.
According to another particular aspect of the invention, a nonlinear function is the sign function or the exponential function or the absolute value function.
According to another particular aspect of the invention, the parametric uncertainties related to the sign of the articular velocity of the robot are uncertainties regarding the dry frictions.
The subject of the invention is also a computer program comprising instructions for the execution of the method of detecting collision according to the invention, when the program is executed by a processor.
The subject of the invention is also a recording medium readable by a processor on which is recorded a program comprising instructions for the execution of the method of detecting collision according to the invention, when the program is executed by a processor.
The subject of the invention is also a command system for a robot comprising a control member for the manipulation of the robot, an interface for the exchange of information regarding the state of the robot and a module for detecting collisions which is adapted for implementing the method according to the invention.
Other characteristics and advantages of the present invention will become more apparent on reading the description which follows in relation to the appended drawings in which:
The method according to the invention receives as input, an item of information E regarding the state of the robot, at any instant or over a given time interval. The item of information E regarding the state of the robot consists of state variables of the robot. These variables can for example take the form of the triplet {q, {dot over (q)}, {umlaut over (q)}} comprising the position q, the velocity {dot over (q)} and the acceleration {umlaut over (q)} for each articulation of the robot. As will be explained in greater detail subsequently, the item of information E regarding the state of the robot can also consist of a single of the three aforementioned state variables or of two of these variables or else of any linear or nonlinear function of one or more of these three variables. In particular but not solely, the possible nonlinear functions are the sign function, the exponential function, the absolute value function or any combination of one or more of these functions. Furthermore the item of information E regarding the state of the robot also comprises a measurement τ of the articular torques of the robot or an item of information making it possible to deduce this measurement, for example a measurement of the motor current on each articulation of the robot.
A first part 100 of the method according to the invention consists in generating a residual r which contains the item of information regarding the events associated with a collision of the robot with its environment.
This first part 100 comprises a first step 101 of measurement τ of the articular torques of the robot, that is to say of the torques measured at the level of each articulation of the robot. As explained hereinabove, this step may, for example, be carried out on the basis of measurements of the motor currents on each articulation.
The first part 100 of the method also comprises a second step 102 of determining a model of the behavior of the robot leading to an estimation {circumflex over (τ)} of the articular torques of the robot.
Various models are possible and known to the person skilled in the art. A particular example of a dynamic model of a rigid serial robot is defined by the following relations:
{circumflex over (τ)}=Â(q){umlaut over (q)}+Ĥ(q,{dot over (q)})+τc (1)
with:
Ĥ(q, {dot over (q)})=Ĉ(q, {dot over (q)}){dot over (q)}+{circumflex over (Q)}(q)+ (2)
q, {dot over (q)}, {umlaut over (q)} are the vectors of the respective estimations or measurements of the articular positions, velocities and accelerations.
{circumflex over (τ)} is a vector of estimates of the articular torques.
τc is a vector of collision torques. The forces applied to the robot's terminal tool can be reflected at the motor level for example by way of the relation:
τc=J(q)TFc
where Fc is the wrench of the exterior forces applied to the robot and J(q) is the Jacobian matrix of the robot.
Â(q) is an estimate of the inertia matrix of the robot.
Ĉ(q, {dot over (q)}) is an estimate of the matrix of the centrifugal and Coriolis terms.
{circumflex over (Q)}(q) is a vector of the estimates of the gravity torques.
is a vector of the estimates of the friction torques. The frictions may, for example, be modeled by a Coulomb model according to the following relation:
τf=Fv{dot over (q)}+Fssign({dot over (q)}) (2′)
where Fv is the coefficient of viscous frictions and Fs the coefficient of dry frictions and sign( ) designates the sign function. The term Fv{dot over (q)} represents an estimate of the viscous frictions. The term Fssign({dot over (q)}) represents an estimate of the dry frictions.
The unknown in equation (1) is the vector of collision torques, step 102 of the method according to the invention therefore consists in estimating the vector of articular torques according to relation (3):
{acute over (τ)}=Â(q){umlaut over (q)}+Ĥ(q, {dot over (q)}) (3)
The dynamic model of the robot exhibited hereinabove is given by way of illustrative example but ought not be interpreted as limiting of the scope of the invention. Indeed, other models may be used, for example, a model similar to that of equation (1) in which one of the two terms A(q){umlaut over (q)} or H(q, {dot over (q)}) is ignored. Generally an estimate of an articular torque of the robot is composed of at least any one term from among the following terms: a term dependent on the articular acceleration, a term dependent on the articular velocity, a term dependent on the articular position, a term dependent on a linear or nonlinear function of any one of the above three terms or of a combination of these terms, for example the sign function, the exponential function or the absolute value function.
The first part 100 of the method according to the invention furthermore comprises a third step 103 of generating the residual r by calculating the difference between the articular torques measurement produced by the first step 101 and the articular torques estimate produced by the second step 102: r=τ−{circumflex over (τ)}
Any scheme known to the person skilled in the art, alternative to that described hereinabove for the first part 100 of the method according to the invention, may be used as replacement insofar as it makes it possible to obtain a residual, that is to say a signal representative of a collision between the robot and its environment. For example, document [5] describes a scheme for generating the residual based on the comparison of momenta rather than articular torques. This scheme can be used as replacement for the first part 100 of the method according to the invention.
On the basis of relations (1) and (3), the residual r can be expressed as the sum of the collision torques τc and of at least one term corresponding to the modeling errors in the models used to estimate the articular torques. The modeling errors comprise on the one hand parametric uncertainties and on the other hand non-parametric uncertainties.
The parametric uncertainties are the bounded errors that may affect the parameters of a model of known structure, for example a model represented by the aforementioned relations (1) and (2). These errors are due to the experimental identification or to the intrinsic variation of the parameters of the model as a function of the operating conditions, for example the variation of the payload, the temperature, the aging.
The non-parametric uncertainties encompass errors which are not related to the model itself but to other phenomena such as the flexibilities of the robot or the friction model used.
In the example given hereinabove where the estimation of the articular torques is obtained with the aid of relation (3), the residual r can be expressed with the aid of the following relations:
r=τ−{circumflex over (τ)}=τ
c
+ΔA(q){umlaut over (q)}+ΔH(q, {dot over (q)}) (4)
ΔA(q)=A(q)−Â(q)
ΔH(q, {dot over (q)})=H(q, {dot over (q)})−Ĥ(q, {dot over (q)})
ΔA(q) et ΔH(q, {dot over (q)}) are the errors between the real values and the estimated values of the terms employed to model the behavior of the robot, in the chosen example this entails the inertia matrix A of the robot and the vector H defined by relation (2).
These errors have a direct impact on the value of the residual which can no longer be considered to be strictly equal to the vector of collision torques (for each articulation). Consequently, even in the absence of collision, the value of the residual is not always zero, thereby posing the problem of the criterion to be applied in order to evaluate the residual and produce a decision regarding the occurrence of a detection.
An aim of the invention is to carry out a filtering of the residual and a formulation of a dynamic detection threshold which makes it possible to circumvent fluctuations related to the errors between the dynamic model of the robot and its real behavior. These processings are grouped together in a second part 200 of the method according to the invention which comprises a step 104 of filtering the residual r so as to obtain a filtered residual rf, a step 105 of on-line formulation of a dynamic detection threshold T as a function of the residual r and of the state of the robot E and finally a step 106 of comparing the filtered residual rf with the detection threshold T in order to formulate a decision D on the presence or the absence of collisions.
Before describing the steps implemented in the second part 200 of the method according to the invention, the following modeling of the residual is introduced for a better understanding of the principle on which the invention is based. Accordingly, we consider the field of Z transforms. Relation (5) gives the modeling of the residual which serves as the basis for the invention.
r
m(z)=τc(z)+Σi=1NSi(z)ei(z)+G0(z)b(z) (5)
The vector of the model of the residual rm(z) is composed of three elements. The first component is the vector of collision torques τc(z) which is the signal that it is sought to detect. The second component Σi=1NSi(z)ei(z) corresponds to the modeling errors, in particular due to parametric uncertainties. The vectors ei(z) represent the information regarding the state of the robot, namely the state variables of the robot such as the acceleration, the velocity, the position or any linear or nonlinear function of one of these variables or of a combination of these variables. The function Si(z) is the transfer function of the modeling errors associated with the item of information ei(z).
Stated otherwise, in the example described hereinabove where the dynamic model of the robot complies with relation (1), it is seen that the parametric uncertainties are dependent on three distinct terms, the articular acceleration {umlaut over (q)}, the articular velocity {dot over (q)} and the sign of the articular velocity sign({dot over (q)}). The model arising from relation (5) may be then written, for this particular example:
r
m(z)=τc(z)+S1(z)e1(z)+S2(z)e2(z)+S3(z)e3(z)+G0(z)b(z) (6)
The terms e1(z), e2(z), e3(z) correspond respectively to the Z transforms of the articular acceleration {umlaut over (q)}, of the articular velocity {dot over (q)} and of the sign of the articular velocity.
The transfer function S1(z) is aimed at modeling the modeling errors related to the acceleration terms in relation (4), namely the term ΔA(q){umlaut over (q)}. This term is particularly significant when the robot is subject to trajectories with large dynamic ranges.
The transfer function S2(z) is aimed at modeling the parametric uncertainties related to the velocity terms in relation (4), stated otherwise the components of the term ΔH(q, {dot over (q)}) which depend on the articular velocity. This term takes into account the velocity-dependent non-linearities, such as the phenomena related to the frictions as well as parametric uncertainties in the Coriolis and centrifugal vectors. With reference to relations (2) and (2′), it may be seen that the velocity dependent terms, in the expression for the dynamic model of the robot, are dependent on the estimates used for the matrix of the centrifugal and Coriolis terms C(q, {dot over (q)}) (relation (2)) and for the friction torques τf (relation (2′)).
Finally the transfer function S3(z) is aimed at modeling the modeling errors related to the terms corresponding to the sign of the velocity in relation (4) thus making it possible to take account of the uncertainties in the dry frictions (see relation (2′)).
The third component of the model of the residual rm(z), namely the term G0(z)b(z) represents a vector of filtered white noise b(z), which makes it possible to take account of the measurement noise and unmodeled dynamic ranges such as the flexibilities which may be present in lightweight serial robots. Such flexibilities are characterized by resonant modes at low frequencies. In the model proposed, according to the invention, by relations (5) and (6), the white noise b(z) impacting the residual is filtered by a low-pass filter G0(z). The term G0(z)b(z) equivalent to this filtering operation corresponds in fact to noise whose frequency spectrum is situated predominantly in the low-frequencies. Indeed, the measurement noise which impacts the model of dynamic evolution of the robot is generally related to low-frequency phenomena such as gravitational effects or low-frequency flexibilities. These errors enter the category of the above-mentioned non-parametric uncertainties. On the contrary, collision phenomena are generally manifested by frequencies covering a wide spectrum and the more the contact, during collision, is characterized by an interaction of large stiffness with the environment, the more the resulting signal comprises high-frequency components. An objective of the method according to the invention therefore consists in distinguishing the high-frequency components, characteristic of a collision, from the low-frequency components, characteristic of measurement noise.
In the first step 104 of the second part 200 of the method, the residual r, obtained on completion of the first part 100 of the method, is filtered by a high-pass filter G0−1(z) so as to render the residual independent of the non-parametric uncertainties related to the low-frequency phenomena, as explained hereinabove. The transfer function of the high-pass filter is, by way of illustration, represented by the term G0−1(z) which is the inverse of the low-pass filter G0(z) used in the model of the residual according to relations (5) and (6). In practice the coefficients of the high-pass filter used are determined with the aid of an adaptive algorithm, based, for example, on the known recursive least squares (RLS) model. The function carried out by such an algorithm consists in calculating the coefficients of the envisaged transfer function, here the high-pass filter, by taking account of the past measurements and of the set of coefficients calculated at the previous instant.
The adaptive filtering of the residual is represented in
Optionally, a further filtering module 203 can be added. This module operates a temporal filtering of each component of the filtered residuals vector rf so as to eliminate the noise that may result from a numerical differentiation. The filtering module 203 may, for example, be embodied through a root mean square calculation. Alternatively, the filtering module 203 may be replaced by the absolute value function.
An exemplary embodiment of the adaptive filtering of the residual r by way of a recursive least squares (RLS) scheme is now described in greater detail.
The well known equations for implementing the RLS algorithm are firstly recalled.
Let θk be the vector of parameters to be estimated, in the present case this entails the vector of coefficients of the filter G0−1 which is learnt in real time. A parameter of the RLS algorithm is the order of the numerators and of the denominators of the transfer function to be estimated, here G0−1(z). The larger the order of the transfer function, the larger its descriptive power and the better the accuracy of representation of the results. On the other hand, this occurs to the detriment of the calculation time which can become prohibitive. The choice of the order is therefore a compromise between accuracy of the model and calculation time.
Let Φk be the vector of input measurements, in the present case this entails the past successive measurements of the residual r.
The equations of the RLS algorithm can be written in the following manner:
Fk is the covariance matrix of the input quantity of the algorithm (here the residual r). λ is the forgetting factor parameter.
Yk is the input quantity of the algorithm, equal to the residual rk in the present case.
εk is the prediction error which corresponds to the criterion to be minimized. In the present case, it is sought to minimize the difference between the residual's current value rk, produced as input of the algorithm, and the result of the filtering, by the transfer function with the estimated coefficients θk-1, of the past values of the residual. The index k represents the index of the current value of a quantity, the indices k-i represent the indices of the past values of a quantity.
Accordingly, we put
Yk=rk
Φk=[rk-1 . . . rk-n
The filtered residual is calculated with the aid of the following relation:
Equations (7),(8),(9) hereinabove are given by way of illustration to explain a possible implementation of the estimation of the coefficients of the filter G0−1. These equations must not be interpreted as limiting, it being understood that the person skilled in the art knows how to implement any other adaptive algorithm or any variant of implementation of the adaptive RLS algorithm so as to arrive at the same result envisaged by the present invention.
In the second step 105 of the second part 200 of the method, a dynamic collision detection threshold T is determined on-line, that is to say in tandem with the displacement of the robot.
The dynamic threshold T is calculated so as to take into account the various modeling errors, such as described hereinabove, affecting the dynamic model of the robot, in particular the parametric and non-parametric uncertainties.
In the example of
Returning to relation (6) and multiplying both sides by the transfer function of the high-pass filter G0−1(z), a model of the filtered residual is obtained:
r
f(z)=G0−1(z)τc(z)+Δ1(z)e1(z)+Δ2(z)e2(z)+Δ3(z)e3(z)+b(z) (7)
With Δi(z)=G0−1(z)Si(z)
According to the model of relation (7), the filtered residual is therefore composed of three elements:
An objective of the second step 105 of the second part 200 of the method according to the invention consists in estimating on-line the transfer functions Δi(z), for each input variable ei(z), and in filtering this variable by a filter reproducing this transfer function. The coefficients of the transfer function Δi(z) are estimated in a recursive manner so as to produce non-stationary coefficients.
Accordingly, a possible scheme consists in using, just as for the filtering of the residual, an adaptive algorithm of the recursive least squares (RLS) type. Any other scheme of the state of the art allowing a recursive estimation of the coefficients is conceivable, in particular a heuristic algorithm, a genetic algorithm, a particle filter, a gradient scheme or any equivalent optimization scheme.
Returning to equations (7),(8) and (9) given hereinabove, we apply in the same manner as for the filtering of the residual, the RLS algorithm with the following parameters for each of the input variables ei(z).
The input quantity of the algorithm Yk is equal to the filtered residual rfk.
The current value of the dynamic threshold TΔi(k) is calculated with the aid of the following relation TΔi(k)=θTkΦk, with θk being the estimate, calculated with the aid of relation (9), of the coefficients of the filter Δi(z).
The adaptive algorithm is therefore applied for each input variable ei(z) and consists as explained hereinabove of a first part 211,212,213 which corresponds to the learning of the coefficients of the transfer function of the filter Δi(z) and in a second part 221,222,223 which corresponds to the filtering itself carried out on the input variable ei(z) so as to obtain a component TΔi of the dynamic threshold.
On output from the adaptive algorithm, the component TΔi is a vector quantity which contains the components associated with the various articulations of the robot.
As already explained hereinabove, optionally, an additional temporal filtering module 231,232,233 can be added to eliminate residual noise in the components TΔi . These filtering modules 231,232,233 may, for example, be embodied through a root mean square calculation. Alternatively, they can be replaced by the absolute value function.
Ultimately, the global dynamic threshold T is obtained by summing the components TΔi obtained for each input variable ei(z) and by adding a static component Tstatic whose value is adjusted as a function of the noise level.
In the example of
T=T
static
+T
Δ1
+T
Δ2
+T
Δ3
The last step 106 of the method according to the invention consists in comparing the filtered residual rf output by step 104 with the global dynamic threshold T and deducing therefrom the existence of a collision if the filtered residual exceeds said threshold. This comparison may be performed for each component of the threshold T corresponding to each articulation of the robot. A collision may be detected if the filtered residual exceeds the dynamic threshold T for at least one articulation. Other variant embodiments of the final detection step 106 are possible such as the triggering of a detection alert only if a predetermined number K among the N articulations of the robot considered satisfies the criterion of comparison between the filtered residual and the dynamic threshold T. Any other variant embodiment making it possible to formulate a decision regarding the occurrence of a collision as a function of the comparison of the vector of filtered residuals and of the vector of dynamic thresholds T is conceivable and will be understood by the person skilled in the art as forming a fully fledged part of the invention.
In
In
The system 400 according to the invention comprises a control member 401 for controlling the trajectory of the arm of the manipulator robot 500, an input/output physical interface 402 between the system 400 and the robot 500 and a module 403 for detecting collisions.
The control member 401 receives as input a high-level control command 410 to drive the robot 500 and transmits as output 411, via the interface 402, information to the robot 500 so as to manipulate it, for example a reference torque. In return, a measurement 412 of the articular torques and positions of the robot 500 is provided to the system 400. This measurement 412 can be taken directly equal to the reference torque 411. Indeed, in the case of DC motors, a reference torque is dispatched to the power amplifiers driving the motors of the robot and slaved in terms of current by an internal feedback loop independent of the main controller. This slaving in terms of current being very fast on the scale of the dynamics of the robot and of the main sampling time, it is in general possible to assume that the torque actually applied to the robot is equal to the reference torque. By virtue of this assumption it is therefore possible to use the reference torque directly as input to the detection algorithm, without needing to explicitly measure the articular torque. Alternatively, if the reference torque 411 is not available, for example because the control member 401 and the module for detecting collisions 403 are implemented in two distinct items of equipment, a measurement 412 of the articular torques of the robot is necessary.
The module for detecting collisions 403 receives the measurement 412 of the articular torques of the robot and establishes a decision 413 regarding the presence or the absence of collision. This decision 413 is for example a binary decision which is thereafter provided to an interface 414 so as to be utilized.
The module for detecting the collisions 403 can be implemented on the basis of hardware and/or software elements. It can notably be implemented in the guise of a computer program comprising instructions for its execution. The computer program can be recorded on a recording medium readable by a processor.
Number | Date | Country | Kind |
---|---|---|---|
1351268 | Feb 2013 | FR | national |
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/EP2014/052394 | 2/7/2014 | WO | 00 |